Non-Intrusive Fish Weight Estimation in Turbid Water Using Deep Learning and Regression Models

Area of Science:

• Aquaculture technology
• Computer vision
• Biometrics

Background:

• Efficient fish management, including feeding and harvesting, is crucial for sustainable aquaculture.
• Accurate fish weight estimation is challenging in turbid underwater environments.
• Existing methods often rely on costly equipment like stereo cameras.

Purpose of the Study:

• To develop a non-intrusive, low-cost 2D computer vision method for estimating Tilapia fish weight.
• To address the challenges of monitoring fish in turbid water conditions.
• To provide a practical alternative to expensive underwater monitoring systems.

Main Methods:

• Utilized a Mask Recurrent-Convolutional Neural Network (Mask R-CNN) for fish detection and pixel dimension extraction.
• Estimated fish depth and converted pixel dimensions to centimeters.
• Employed regression learning models (linear regression, random forest, support vector regression) for weight estimation.

Main Results:

• Achieved a Mean Absolute Error (MAE) of 42.54 g for weight estimation.
• Obtained an R-squared (R2) value of 0.70, indicating good model fit.
• Reported an average weight error of 30.30 (±23.09) grams in turbid water.

Conclusions:

• The proposed 2D computer vision framework offers a practical and cost-effective solution for underwater Tilapia weight estimation.
• The method demonstrates feasibility in challenging turbid water environments.
• This technology can support efficient aquaculture management through improved fish monitoring.